NEET UG · 10 papers · 2017–2026
How we analysed ten years of NEET, and how far to trust it
A detailed map of what NEET tests, and how much you can and cannot infer from it: where the questions come from, how we tagged them, and how the patterns held up on papers they had not seen.
- Official papers
- 10
- 2017–2026; the cancelled 2026 paper left out
- Questions tagged
- 1,875
- 5 dropped by NTA left out
- Chapters for half the paper
- 31
- 61 for 80%, of 94 asked
- Exact idea repeated
- 41%
- of 2026 questions; the topic repeated 99%
Key findings
- 1
NEET is broad: the top 10 chapters hold only 21% of questions, and it takes 61 of 94 chapters to reach 80%.
- 2
In 2026, 99% of questions came from a topic asked before, but only 41% repeated the exact idea. Topics come back; questions rarely do.
- 3
History beats guessing, but weighting it adds little: the top 20 topics held 22% of the next paper with recency weighting, 21% with a plain ten-year count, and 10% for a random pick.
- 4
Topic chances are honest: on average the chance we gave was within 6 points of what happened.
- 5
Leaving out the cancelled 2026 paper barely matters: 19 of the top 20 chapters stay the same and no chapter moves by more than 0.5 questions.
Sources
What went in
Only official NEET UG question papers and final answer keys. No coaching material, no memory-based papers.
- Papers. One official paper per year, 2017, 2018, 2019, 2020, 2021, 2022, 2023, 2024, 2025, 2026. Where NEET set several code versions, they hold the same questions, so one is enough.
- The cancelled paper. The first 2026 paper was cancelled and set again. We count only the re-exam, and check below that this choice does not move the results.
- Dropped questions. 5 questions NTA withdrew in the final key are left out. From 2021 to 2024 the paper had 200 questions of which 180 were answered; per-paper figures are scaled to 180.
- Syllabus. Every chapter was checked against the official NEET UG 2026 syllabus. Dropped chapters keep their history but get no weight in plans.
- Tagging. Each question was tagged to an NCERT chapter, a topic and the single idea it tests, plus its format, skill and difficulty. The tagging was done by AI models and checked by AI and scripts, as set out below.
| Year | Paper | Printed | Tagged | New idea |
|---|---|---|---|---|
| 2017 | Code A | 180 | 178 | – |
| 2018 | Code AA | 180 | 180 | 95% |
| 2019 | Code P1 | 180 | 180 | 89% |
| 2020 | Code E1 | 180 | 180 | 84% |
| 2021 | Code O1 | 200 | 200 | 86% |
| 2022 | Code R1 | 200 | 199 | 66% |
| 2023 | Code G6 | 200 | 199 | 64% |
| 2024 | Code T2 | 200 | 200 | 59% |
| 2025 | Code 45 | 180 | 180 | 62% |
| 2026 | Cancelled, left out | 180 | – | – |
| 2026 | Re-exam | 180 | 179 | 59% |
Printed 1,880 questions across kept papers; 1,875 tagged. New idea: share of the year's questions testing an idea not asked in any earlier paper.
Tagging
How each question got its chapter, topic and idea
Every step below was done by AI models and checked by AI and scripts. Here is what was checked, what it found, and what was never measured.
- 1 · Transcribe2,060questions
AI models read 11 official papers and answer keys into text. A script checked numbering, counts, four options each, and every answer against the official key. Figures had a separate page-by-page pass.
- 2 · Tag1 ideaper question
AI tagging agents placed each question on a chapter, a topic and one idea from Lumi's syllabus map, rated how well it fitted, and added format, skill and difficulty.
- 3 · Spot-check218/220re-checks agreed
The AI coordinating the work re-checked 5 random tags in each of 44 batches, ready to redo any batch with more than 1 in 5 wrong. It disagreed 2 times; both were corrected.
- 4 · Chapter audit42chapter tags fixed
All 2,060 questions were read again for their chapter. 3 of the 1,520 from 2017 to 2024 were wrong; most fixes were Biology in the newest papers.
- 5 · Fill the gaps51ideas added
Where no idea in the map fitted, the missing idea was added and the question tagged again, with 228 topic and chapter corrections, each with a written reason.
How well tags fit now
- 1,451 good fit: the idea is exactly what the question tests
- 424 looser fit: right area, idea a little broader or narrower
Fit is the tagger’s own rating, not an outside check. No question is left on a poor fit.
Not measured
No person re-tagged a sample on their own, and we have not measured how often two taggers would agree. The spot-checks were made by the AI that coordinated the tagging, so they show the work is consistent, not that it is independently right.
Treat chapter tags as close to certain, and topic and idea tags as careful judgement calls. An independent human check of a random sample is the next step.
What “same topic” and “same idea” mean
Real questions from the official papers. A topic is a part of a chapter; an idea is the one fact or skill a question turns on. Repeats on this site count ideas, which is why they are rarer than repeated topics.
- Same ideaDNA Double Helix
NEET 2017 · A · Q100
The final proof for DNA as the genetic material came from the experiments of :
Idea: Hershey-Chase Experiment
NEET 2023 · G6 · Q118
Unequivocal proof that DNA is the genetic material was first proposed by
Idea: Hershey-Chase Experiment
Six years apart, the same fact: the experiment that proved DNA is the genetic material. Only the wording changed.
- Same topic, different ideaDNA Double Helix
NEET 2017 · A · Q100
The final proof for DNA as the genetic material came from the experiments of :
Idea: Hershey-Chase Experiment
NEET 2021 · O1 · Q172
If Adenine makes 30% of the DNA molecule, what will be the percentage of Thymine, Guanine and Cytosine in it ?
Idea: Chargaff's Rules
Same topic, the DNA double helix, but one asks for an experiment and the other applies Chargaff's rule to base percentages.
- Same ideaChordate Classes
NEET 2018 · AA · Q148
Identify the vertebrate group of animals characterized by crop and gizzard in its digestive system.
Idea: Aves
NEET 2022 · R1 · Q155
In which of the following animals, digestive tract has additional chambers like crop and gizzard ?
Idea: Aves
Both ask which animals have a crop and gizzard. Knowing the bird digestive tract answers both.
- Same topic, different ideaChordate Classes
NEET 2017 · A · Q141
Which among these is the correct combination of aquatic mammals ?
Idea: Mammalia
NEET 2018 · AA · Q148
Identify the vertebrate group of animals characterized by crop and gizzard in its digestive system.
Idea: Aves
Both sit on chordate classes, but one is about aquatic mammals and the other about birds: different facts to learn.
- Same ideaMeiotic Prophase I
NEET 2022 · R1 · Q128
The appearance of recombination nodules on homologous chromosomes during meiosis characterizes :
Idea: Recombination Nodule Identification
NEET 2023 · G6 · Q115
The process of appearance of recombination nodules occurs at which sub stage of prophase I in meiosis?
Idea: Recombination Nodule Identification
In back-to-back years: at which stage of prophase I recombination nodules appear.
- Same topic, different ideaMeiotic Prophase I
NEET 2020 · E1 · Q86
Dissolution of the synaptonemal complex occurs during :
Idea: Initiation of Synaptonemal Complex Dissolution
NEET 2021 · O1 · Q176
Which stage of meiotic prophase shows terminalisation of chiasmata as its distinctive feature ?
Idea: Terminalisation
Both are meiotic prophase I, but one is when the synaptonemal complex dissolves and the other when chiasmata terminalise.
- Same ideaRestriction Enzymes
NEET 2019 · P1 · Q122
DNA precipitation out of a mixture of biomolecules can be achieved by treatment with :
Idea: DNA Precipitation with Chilled Ethanol
NEET 2021 · O1 · Q101
During the purification process for recombinant DNA technology, addition of chilled ethanol precipitates out :
Idea: DNA Precipitation with Chilled Ethanol
The same step of recombinant DNA work, precipitating DNA with chilled ethanol, asked two years apart.
- Same topic, different ideaIdeal Gas Law
NEET 2018 · AA · Q86
The correction factor 'a' to the ideal gas equation corresponds to
Idea: van der Waals Equation
NEET 2022 · R1 · Q60
Which one is not correct mathematical equation for Dalton's Law of partial pressure ? Here p = total pressure of gaseous mixture
Idea: Dalton's Law of Partial Pressures
Both are on the gas laws, but one asks what the van der Waals constant a corrects for and the other about Dalton's law of partial pressures.
Concentration
There is no short list of chapters
The top 31 chapters cover half the questions; 61 cover 80%. That is 65% of all chapters asked. Topics are a little more concentrated: 56 of 204 cover half.
- Chapters (94)
- Topics (204)
- If every one were equal
Repetition
Chapters and topics repeat. The exact idea mostly does not
For each year, the share of questions whose chapter, topic or exact idea had appeared in any earlier paper.
- Chapter asked before
- Topic asked before
- Exact idea asked before
Backtest
Tested on papers it had not seen
For each paper from 2022 to 2026, we ranked chapters, topics and ideas using only the papers before it, then checked the ranking against that year's real paper, next to three simpler ways of picking.
History against guessing
21% vs 10%
Picking the 20 topics asked most in the papers before, with no weighting at all, held 21% of the next paper. Twenty topics picked at random held 10%. Most of what past papers can tell you comes from simply counting them.
Weighting against counting
22% vs 21%
Letting recent papers count more added 0.7 points of the paper for the top 20 topics, about one question. For the top 20 chapters it was 36% against 35%, and 49% against 49% for the top 30. For exact ideas the plain count did better at top 200. A real but small edge, not a prediction.
Four ways to pick, side by side
Share of the next paper’s questions held by each method’s top picks, averaged over the 5 papers from 2022 to 2026. A random pick is the expected share for that many topics or chapters chosen blind. A gap under one question of a 180-question paper counts as a tie.
- Top 10 chaptersAhead · +2.2 Q
- Weighted estimate21%
- Ten-year count19%
- Last year only19%
- Random pick11%
- Top 20 chaptersAhead · +1.6 Q
- Weighted estimate36%
- Ten-year count35%
- Last year only35%
- Random pick21%
- Top 30 chaptersTie · +0.7 Q
- Weighted estimate49%
- Ten-year count49%
- Last year only48%
- Random pick32%
- Top 20 topicsAhead · +1.3 Q
- Weighted estimate22%
- Ten-year count21%
- Last year only20%
- Random pick10%
- Top 40 topicsAhead · +2.0 Q
- Weighted estimate38%
- Ten-year count37%
- Last year only34%
- Random pick20%
- Top 60 topicsTie · +0.9 Q
- Weighted estimate51%
- Ten-year count51%
- Last year only48%
- Random pick30%
- Top 50 ideasTie · −0.9 Q
- Weighted estimate4%
- Ten-year count5%
- Last year only2%
- Random pick2%
- Top 100 ideasTie · +0.9 Q
- Weighted estimate7%
- Ten-year count6%
- Last year only4%
- Random pick4%
- Top 200 ideasBehind · −2.7 Q
- Weighted estimate10%
- Ten-year count12%
- Last year only9%
- Random pick7%
| List | Weighted estimate | Ten-year count | Last year only | Random pick | Weighting vs count |
|---|---|---|---|---|---|
| Top 10 chapters | 21% | 19% | 19% | 11% | Ahead · +2.2 Q |
| Top 20 chapters | 36% | 35% | 35% | 21% | Ahead · +1.6 Q |
| Top 30 chapters | 49% | 49% | 48% | 32% | Tie · +0.7 Q |
| Top 20 topics | 22% | 21% | 20% | 10% | Ahead · +1.3 Q |
| Top 40 topics | 38% | 37% | 34% | 20% | Ahead · +2.0 Q |
| Top 60 topics | 51% | 51% | 48% | 30% | Tie · +0.9 Q |
| Top 50 ideas | 4% | 5% | 2% | 2% | Tie · −0.9 Q |
| Top 100 ideas | 7% | 6% | 4% | 4% | Tie · +0.9 Q |
| Top 200 ideas | 10% | 12% | 9% | 7% | Behind · −2.7 Q |
Paper by paper
Each line is one graded paper. Across these 15 comparisons the weighting was ahead of the plain count 7 times, tied 6 and behind 2.
Top 20 topics
Against the ten-year count: ahead in 2, tied in 2, behind in 1 of 5 papers.
- 2022Tie · 0.0 Q
- 2023Ahead · +5.4 Q
- 2024Behind · −2.7 Q
- 2025Ahead · +4.0 Q
- 2026Tie · 0.0 Q
Top 20 chapters
Against the ten-year count: ahead in 3, tied in 2, behind in 0 of 5 papers.
- 2022Ahead · +1.8 Q
- 2023Ahead · +2.7 Q
- 2024Tie · 0.0 Q
- 2025Tie · 0.0 Q
- 2026Ahead · +4.0 Q
Top 30 chapters
Against the ten-year count: ahead in 2, tied in 2, behind in 1 of 5 papers.
- 2022Ahead · +3.6 Q
- 2023Ahead · +1.8 Q
- 2024Tie · +0.9 Q
- 2025Tie · +1.0 Q
- 2026Behind · −4.0 Q
- Weighted estimate
- Ten-year count
- Last year only
- Random pick
One caution. The weighting’s settings (how fast older papers fade, a half-life of three years) were chosen by trying several on these same 5 papers. That makes its small lead look a little better than it would on papers it has never met.
The full numbers
Hit rate: share of the top picks that were asked. Coverage: share of the paper's questions they held. Averaged over the 5 papers from 2022 to 2026.
- Top 50 ideas2.3× random
Hit rate 14% (random 7%) · coverage 4% (random 2%)
- Top 100 ideas1.9× random
Hit rate 12% (random 7%) · coverage 7% (random 4%)
- Top 200 ideas1.4× random
Hit rate 10% (random 7%) · coverage 10% (random 7%)
- Top 10 chapters1.9× random
Hit rate 100% (random 86%) · coverage 21% (random 11%)
- Top 20 chapters1.7× random
Hit rate 98% (random 86%) · coverage 36% (random 21%)
- Top 30 chapters1.5× random
Hit rate 98% (random 86%) · coverage 49% (random 32%)
- Top 20 topics2.2× random
Hit rate 94% (random 60%) · coverage 22% (random 10%)
- Top 40 topics1.9× random
Hit rate 91% (random 60%) · coverage 38% (random 20%)
- Top 60 topics1.7× random
Hit rate 85% (random 60%) · coverage 51% (random 30%)
Questions per chapter: our weighting was off by 0.79 questions per chapter on average, against 1.02 for a flat guess, and its ranking agreed with the real paper at a rank correlation of 0.63. Knowing each year's syllabus changes in advance would have cut the error only to 0.70.
Calibration
When we say 40%, does it happen 40% of the time?
We grouped every past forecast by the chance it gave, then counted how often the thing really came. Dots on the dashed line mean the chances are honest.
- Topics
- Exact ideas, before correction
- Perfect: said = happened
Bigger dots stand for more cases.
Sensitivity
Does the cancelled paper change anything?
We ran the chapter weighting twice: with the cancelled paper and without it.
- Rank agreement
- 0.99
- 1.00 means the same order
- Top 20 the same
- 19 of 20
- chapters in both lists
- Biggest shift
- 0.5 Q
- Equilibrium
The two rankings are almost identical. Leaving the cancelled paper out is a safe choice, and nothing on these pages depends on it.
Priority bands
How a chapter gets its band
Chapters in the 2026 syllabus are ranked by their ten-year average questions a paper (history, not the model). Going down that list, the chapters that together held the first half of all questions are Must do; the next ones, up to 80%, are Should do; the rest are Lower priority.
Bands use history rather than the weighted estimate because the backtest shows weighting adds only about one question a paper over plain counting, and history is the part we can state as fact. Study hours on chapter pages are what the study plan gives each chapter: its weighted estimate, adjusted for how hard its questions run and how many other chapters build on it. They are a planning model, not timings of real students.
- Must do28chapters · First 50% of questions
- Should do26chapters · 50% to 80%
- Lower priority26chapters · Last 20%
Limits
What this cannot tell you
- Tagging was done and checked by AI. No person re-tagged a sample independently, so there is no measured agreement rate. Chapter tags were audited in full and are close to certain. Topic and idea tags are judgement calls, and some will be wrong.
- Every question sits on a topic and an idea. All 1,875 questions carry a chapter, a topic and an idea, so chapter, topic and idea figures are counted over the same questions. Where Lumi’s idea list had no close match, the missing ideas were added to it and the question tagged again. Some tags are still a looser fit: the right area, but an idea a little broader or narrower than the one tested.
- Early years have little history. “New idea” in 2018 means new since 2017 only. The share of new ideas looks higher in early years partly for that reason.
- The syllabus changed in 2024. Chapters were dropped and a few added. Patterns from before 2024 count for less in chapters that changed.
- Ten papers is a short history. One surprise year moves a chapter's average. Treat weightage as a guide to effort, never as a forecast of the paper.
Questions